Author: Uri Blass
Date: 01:43:52 08/22/03
Go up one level in this thread
On August 22, 2003 at 04:15:30, Gian-Carlo Pascutto wrote: >On August 21, 2003 at 17:33:57, Uri Blass wrote: > >>We can only say that we did not see a proof that Noomens book is superior. > >That is true. > >It's hard to draw a conclusion from those games. The book doesn't >make hundreds of points of difference and the programs aren't all >that far apart either. I agree with you on that. So whoever wins >has a lot to do with luck. > >You used the word 'evidence' earlier but that's severely misplaced. >It becomes evidence when it's statistically significant. Before that, >I wouldn't put any faith in it. > >-- >GCP Maybe there is a misunderstanding. I did not mean to say that it is statistically significant when I said evidence. The point is that even if the data is not statistically significant it suggests something and I do not think that it is right to ignore the data only because of the fact that it is not statistically significant. I know that there are programmers who do not do it see http://www.f11.parsimony.net/forum16635/messages/52921.htm for comparison that he used to decide that the change is not productive see http://www.innconx.com/~wildcat/279.html http://www.innconx.com/~wildcat/282.html 282 scored one point more than 279 but the programmer decided not to accept the change. Note that I do not use a lot of testing like him but if there is a change that I am not sure if to do it and a beta tester tell me about these results then I accept the change. Of course the difference is too small to get significant results but there is a basis to give bigger probability to the event that 282 is better and not to the event that 279 is better. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.